A5077078493- Particle Detector Development and Performance
- Radiation Detection and Scintillator Technologies
- Advanced Neural Network Applications
- Scientific Computing and Data Management
- Nuclear reactor physics and engineering
- Distributed and Parallel Computing Systems
- Radiation Effects in Electronics
- Parallel Computing and Optimization Techniques
- Software System Performance and Reliability
- Digital Media Forensic Detection
- Simulation Techniques and Applications
- CCD and CMOS Imaging Sensors
- Generative Adversarial Networks and Image Synthesis
- Photonic and Optical Devices
- Machine Learning in Materials Science
- Electrostatic Discharge in Electronics
- Medical Imaging Techniques and Applications
- Particle accelerators and beam dynamics
- Muon and positron interactions and applications
- Model Reduction and Neural Networks
- Optical Network Technologies
- Advancements in PLL and VCO Technologies
- Astrophysics and Cosmic Phenomena
- Neural Networks and Applications
- Magnetic confinement fusion research
Fermi National Accelerator Laboratory
2021-2024
Columbia University
2004
We describe a method for precisely regulating the gradient magnet power supply at Fermilab Booster accelerator complex using neural network trained via reinforcement learning. demonstrate preliminary results by training surrogate machine-learning model on real data to emulate environment, and this in turn train its regulation task. additionally show how networks be deployed control purposes may compiled execute field-programmable gate arrays. This capability is important operational...
We develop an end-to-end workflow for the training and implementation of co-designed neural networks (NNs) efficient field-programmable gate array (FPGA) hardware. Our approach leverages Hessian-aware quantization NNs, Quantized Open Neural Network Exchange intermediate representation, hls4ml tool flow transpiling NNs into FPGA firmware. This makes NN implementations in hardware accessible to nonexperts a single open sourced that can be deployed real-time machine-learning applications wide...
We present extensions to the Open Neural Network Exchange (ONNX) intermediate representation format represent arbitrary-precision quantized neural networks. first introduce support for low precision quantization in existing ONNX-based formats by leveraging integer clipping, resulting two new backward-compatible variants: operator with clipping and quantize-clip-dequantize (QCDQ) format. then a novel higher-level ONNX called (QONNX) that introduces three operators -- Quant, BipolarQuant,...
This paper presents the algorithms and architecture proposed for upgrade of level 1 calorimeter trigger D0 experiment. We describe digital-signal-processing algorithm applied to individual tower signals physics that process complete array towers. investigate performance these justify our choices. present hardware system designed analyze 2560 samples construct primitives in /spl sim/3 mu/s at a rate 7.57 MHz. give detailed description two prototype boards are being built: analog-to-digital...
We develop an end-to-end workflow for the training and implementation of co-designed neural networks (NNs) efficient field-programmable gate array (FPGA) application-specific integrated circuit (ASIC) hardware. Our approach leverages Hessian-aware quantization (HAWQ) NNs, Quantized Open Neural Network Exchange (QONNX) intermediate representation, hls4ml tool flow transpiling NNs into FPGA ASIC firmware. This makes NN implementations in hardware accessible to nonexperts, a single open-sourced...
existing operations-critical functions of the BLM system. This paper will focus on evolution architectures, which provided high-frequency, low-latency collection synchronized data streams to make real-time inferences. The ML models, used for learning both local and global machine signatures producing high quality inferences based raw loss measurements, only be discussed at a high-level.
In this community review report, we discuss applications and techniques for fast machine learning (ML) in science -- the concept of integrating power ML methods into real-time experimental data processing loop to accelerate scientific discovery. The material report builds on two workshops held by Fast Science covers three main areas: across a number domains; training implementing performant resource-efficient algorithms; computing architectures, platforms, technologies deploying these...
Extracting low-energy signals from LArTPC detectors is useful, for example, detecting supernova events or calibrating the energy scale with argon-39. However, it difficult to efficiently extract because of noise. We propose using a 1DCNN select wire traces that have signal. This suppresses background while still being efficient then followed by 1D autoencoder denoise traces. At point signal waveform can be cleanly extracted. In order make this processing efficient, we implement two networks...
model may be producible to help de-blend losses between machines. Work is underway as part of the Fermilab Real-time Edge AI for Distributed Systems Project (READS) develop a ML empowered system that collects streamed BLM data and additional machine readings infer in real-time, which generated beam loss.
This study focuses on implementing a real-time control system for particle accelerator facility that performs high energy physics experiments. A critical operating parameter in this is beam loss, which the fraction of particles deviating from accelerated proton into cascade secondary particles. Accelerators employ large number sensors to monitor loss. The data these monitored by human operators who predict relative contribution different sub-systems Using information, they engage...
This work enables active control and suppression of MHD instabilities in magnetic confinement fusion devices such as the Tokamak with a feedback system using high speed cameras deep learning on frame grabber FPGAs.
We introduce a novel Proximal Policy Optimization (PPO) algorithm aimed at addressing the challenge of maintaining uniform proton beam intensity delivery in Muon to Electron Conversion Experiment (Mu2e) Fermi National Accelerator Laboratory (Fermilab). Our primary objective is regulate spill process ensure consistent profile, with ultimate goal creating an automated controller capable providing real-time feedback and calibration Spill Regulation System (SRS) parameters on millisecond...
using a memory look-up table addressed with nine to ten most significant non-zero bits. At 200 MHz internal clock, our demo core reaches throughput of M pairs/s/core, faster than typical 2 GHz micro-processor by about factor 10. Temperature and power consumption FPGAs were also lower those micro-processors. Fast convenient, can serve as alternatives time-consuming micro-processors for space charge simulation.